mict-zhaw commited on
Commit
4e7dd25
1 Parent(s): 3cf8d44

Rename and reorganize

Browse files
Files changed (1) hide show
  1. chall.py +88 -76
chall.py CHANGED
@@ -8,6 +8,8 @@ import soundfile as sf
8
  import importlib.util
9
 
10
 
 
 
11
  _DESCRIPTION = "tbd"
12
  _CITATION = "tbd"
13
 
@@ -130,7 +132,7 @@ class Chall(GeneratorBasedBuilder):
130
  "text": Value("string"),
131
  }
132
  ),
133
- "audio": Audio(sampling_rate=16_000)
134
  })
135
  else:
136
  features = Features({
@@ -168,7 +170,7 @@ class Chall(GeneratorBasedBuilder):
168
  ),
169
  }
170
  ),
171
- "audio": Audio(sampling_rate=16_000)
172
  })
173
 
174
  return DatasetInfo(
@@ -214,26 +216,58 @@ class Chall(GeneratorBasedBuilder):
214
  # ),
215
  ]
216
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
217
  @staticmethod
218
- def _generate_examples_segments(audio_id: str, audio_file_path: str, data: dict, transcript: dict):
219
  """
 
 
 
220
 
221
- :param audio_id:
222
- :param audio_file_path:
223
- :param data:
224
- :param transcript:
225
- :return:
226
  """
227
 
228
- data["speakers"] = transcript.get("speakers", [])
229
- data["segments"] = transcript.get("segments", [])
 
230
 
231
- audio, samplerate = sf.read(audio_file_path)
232
- data["audio"] = {"path": audio_file_path, "array": audio, "sampling_rate": samplerate}
233
 
234
- yield audio_id, data
 
235
 
236
- def _generate_examples_utterances(self, audio_id: str, audio_file_path: str, data: dict, transcript: dict):
237
  """
238
  Generates examples from audio segments based on the transcript provided. Each segment is processed to produce
239
  an utterance which includes the audio slice and metadata.
@@ -242,56 +276,58 @@ class Chall(GeneratorBasedBuilder):
242
  :param audio_file_path: The filesystem path to the audio file.
243
  :param data: A dictionary containing the segments to be processed
244
  :param transcript: A dictionary containing transcript details with segments of spoken words.
245
- :return: Yields a tuple of audio id and it's corresponding data
246
  """
247
 
248
  segments = transcript.get("segments", [])
 
249
 
250
- if self.config.max_pause_length:
251
- segments = self._remove_to_long_pauses(segments)
252
-
253
- if self.config.max_chunk_length is not None:
254
- segments = self._split_long_utterances(segments)
255
-
256
- if self.config.remove_trailing_pauses:
257
- self._remove_trailing_pauses_in_segments(segments)
258
-
259
- if self.config.max_pause_length is not None:
260
- segments = self._filter_utterances_by_duration(segments, self.config.min_chunk_length, self.config.max_chunk_length)
261
 
262
- for segment_i, segment in enumerate(segments):
 
263
 
264
- id_ = f"{audio_id}_{str(segment_i).rjust(3, '0')}"
 
 
265
 
266
- data["audio_id"] = id_
267
- data["speaker_id"] = segment["speaker"]
268
- data["words"] = segment["words"]
269
 
270
- track = sf.SoundFile(audio_file_path)
 
 
 
271
 
272
- can_seek = track.seekable()
273
- if not can_seek:
274
- raise ValueError("Not compatible with seeking")
275
 
276
- sr = track.samplerate
277
- start_time = segment["words"][0]["start"]
278
- end_time = segment["words"][-1]["end"]
279
 
280
- start_frame = int(sr * start_time)
281
- frames_to_read = int(sr * (end_time - start_time))
 
 
 
 
 
 
282
 
283
- # Seek to the start frame
284
- track.seek(start_frame)
285
 
286
- # Read the desired frames
287
- audio = track.read(frames_to_read)
288
 
289
- data["audio"] = {"path": audio_file_path, "array": audio, "sampling_rate": sr}
 
290
 
291
- yield id_, data
292
 
293
  @staticmethod
294
- def _remove_trailing_pauses_in_segments(segments: List[dict]) -> None:
295
  """
296
  Removes pauses at the end/start of utterances in each segment to eliminate pauses between segments.
297
 
@@ -310,10 +346,11 @@ class Chall(GeneratorBasedBuilder):
310
  # Remove segment if no words left
311
  if not segment["words"]:
312
  segments.remove(segment)
 
313
 
314
- def _remove_to_long_pauses(self, segments: List[dict]) -> List[dict]:
315
  """
316
- Remove to long pauses in a segment by splitting the segment in two segments and removing the filled pause.
317
 
318
  Example (assuming (...) is longer than max_pause_length):
319
  [["Hello", "(...)", "World!"]] --> [["Hello"], ["World!"]]
@@ -339,7 +376,7 @@ class Chall(GeneratorBasedBuilder):
339
  return split_segments
340
 
341
  @staticmethod
342
- def _filter_utterances_by_duration(segments: List[dict], min_duration: float = None, max_duration: float = None, ):
343
  """
344
  Removes segments with invalid duration
345
  :param min_duration: The minimum duration allowed for a segment.
@@ -357,7 +394,7 @@ class Chall(GeneratorBasedBuilder):
357
 
358
  return filtered_segments
359
 
360
- def _split_long_utterances(self, segments: List[dict]) -> List[dict]:
361
  """
362
  Splits segments into smaller chunks if their duration exceeds the maximum chunk length specified in the config.
363
 
@@ -406,29 +443,4 @@ class Chall(GeneratorBasedBuilder):
406
 
407
  return list_of_chunks
408
 
409
- def _generate_examples(self, filepath, metafile):
410
- """
411
- This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.
412
- :param filepath: The path where the data is located.
413
- :param metafile: The metafile describing the chall data
414
- :return:
415
- """
416
-
417
- logger.info("generating examples from = %s", filepath)
418
-
419
- with open(metafile, 'r') as file:
420
- for line in file:
421
- data = json.loads(line)
422
-
423
- # load transcript
424
- transcript_file = os.path.join(filepath, data["transcript_file"])
425
- with open(transcript_file, 'r') as transcript:
426
- transcript = json.load(transcript)
427
-
428
- audio_id = data['audio_id']
429
- audio_file_path = os.path.join(filepath, data["audio_file"])
430
 
431
- if self.config.split_segments:
432
- yield from self._generate_examples_utterances(audio_id, str(audio_file_path), data, transcript)
433
- else:
434
- yield from self._generate_examples_segments(audio_id, str(audio_file_path), data, transcript)
 
8
  import importlib.util
9
 
10
 
11
+ _SAMPLE_RATE = 16000
12
+
13
  _DESCRIPTION = "tbd"
14
  _CITATION = "tbd"
15
 
 
132
  "text": Value("string"),
133
  }
134
  ),
135
+ "audio": Audio(sampling_rate=_SAMPLE_RATE)
136
  })
137
  else:
138
  features = Features({
 
170
  ),
171
  }
172
  ),
173
+ "audio": Audio(sampling_rate=_SAMPLE_RATE)
174
  })
175
 
176
  return DatasetInfo(
 
216
  # ),
217
  ]
218
 
219
+ def _generate_examples(self, filepath, metafile):
220
+ """
221
+ This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.
222
+ :param filepath: The path where the data is located.
223
+ :param metafile: The metafile describing the chall data
224
+ :return:
225
+ """
226
+
227
+ logger.info("generating examples from = %s", filepath)
228
+
229
+ with open(metafile, 'r') as file:
230
+ for line in file:
231
+ data = json.loads(line)
232
+
233
+ # load transcript
234
+ transcript_file = os.path.join(filepath, data["transcript_file"])
235
+ with open(transcript_file, 'r') as transcript:
236
+ transcript = json.load(transcript)
237
+
238
+ audio_id = data['audio_id']
239
+ audio_file_path = os.path.join(filepath, data["audio_file"])
240
+
241
+ if self.config.split_segments:
242
+ yield from self._generate_utterance_examples(audio_id, str(audio_file_path), data, transcript)
243
+ else:
244
+ yield from self._generate_transcript_examples(audio_id, str(audio_file_path), data, transcript)
245
+
246
  @staticmethod
247
+ def _generate_transcript_examples(audio_id: str, audio_file_path: str, data: dict, transcript: dict):
248
  """
249
+ Generates examples based on the entire audio file and its associated transcript metadata. This method reads the
250
+ entire audio file, extracts speaker and segment information from the transcript, and packages these along with
251
+ the audio data into a dictionary that is then yielded.
252
 
253
+ :param audio_id: A unique identifier for the audio file.
254
+ :param audio_file_path: The file system path to the audio file.
255
+ :param data: A dictionary of the metadata.
256
+ :param transcript: A dictionary containing details of the transcript, including speakers and segments.
257
+ :return: Yields a tuple containing the audio ID and the enriched transcript dictionary.
258
  """
259
 
260
+ transcript_data = data.copy() # Create a fresh copy of data to ensure no side effects
261
+ transcript_data["speakers"] = transcript.get("speakers", [])
262
+ transcript_data["segments"] = transcript.get("segments", [])
263
 
264
+ with sf.SoundFile(audio_file_path) as audio_file:
265
+ audio = audio_file.read(dtype='float32')
266
 
267
+ transcript_data["audio"] = {"path": audio_file_path, "array": audio, "sampling_rate": _SAMPLE_RATE}
268
+ yield audio_id, transcript_data
269
 
270
+ def _generate_utterance_examples(self, audio_id: str, audio_file_path: str, data: dict, transcript: dict):
271
  """
272
  Generates examples from audio segments based on the transcript provided. Each segment is processed to produce
273
  an utterance which includes the audio slice and metadata.
 
276
  :param audio_file_path: The filesystem path to the audio file.
277
  :param data: A dictionary containing the segments to be processed
278
  :param transcript: A dictionary containing transcript details with segments of spoken words.
279
+ :return: Yields a tuple containing the audio ID and the enriched utterance dictionary.
280
  """
281
 
282
  segments = transcript.get("segments", [])
283
+ segments = self._process_segments(segments)
284
 
285
+ with sf.SoundFile(audio_file_path) as track:
 
 
 
 
 
 
 
 
 
 
286
 
287
+ if not track.seekable():
288
+ raise ValueError("Audio file is not seekable.")
289
 
290
+ for segment_i, segment in enumerate(segments):
291
+ segment_data = data.copy() # Create a fresh copy of data for each segment
292
+ segment_id = f"{audio_id}_{str(segment_i).rjust(3, '0')}"
293
 
294
+ segment_data["audio_id"] = segment_id
295
+ segment_data["speaker_id"] = segment["speaker"]
296
+ segment_data["words"] = segment["words"]
297
 
298
+ start_time = segment["words"][0]["start"]
299
+ end_time = segment["words"][-1]["end"]
300
+ start_frame = int(_SAMPLE_RATE * start_time)
301
+ frames_to_read = int(_SAMPLE_RATE * (end_time - start_time))
302
 
303
+ track.seek(start_frame)
304
+ audio = track.read(frames_to_read)
305
+ segment_data["audio"] = {"path": audio_file_path, "array": audio, "sampling_rate": _SAMPLE_RATE}
306
 
307
+ yield segment_id, segment_data
 
 
308
 
309
+ def _process_segments(self, segments):
310
+ """
311
+ Processes the list of segments based on configured rules.
312
+ :param segments: A list of segment dictionaries
313
+ :return: A list of processed segment dictionaries after applying all the filtering and splitting rules.
314
+ """
315
+ if self.config.max_pause_length:
316
+ segments = self._split_and_remove_long_pauses(segments)
317
 
318
+ if self.config.max_chunk_length is not None:
319
+ segments = self._split_long_segments(segments)
320
 
321
+ if self.config.remove_trailing_pauses:
322
+ segments = self._remove_trailing_pauses(segments)
323
 
324
+ if self.config.max_pause_length is not None:
325
+ segments = self._filter_segments_by_duration(segments, self.config.min_chunk_length, self.config.max_chunk_length)
326
 
327
+ return segments
328
 
329
  @staticmethod
330
+ def _remove_trailing_pauses(segments: List[dict]) -> List[dict]:
331
  """
332
  Removes pauses at the end/start of utterances in each segment to eliminate pauses between segments.
333
 
 
346
  # Remove segment if no words left
347
  if not segment["words"]:
348
  segments.remove(segment)
349
+ return segments
350
 
351
+ def _split_and_remove_long_pauses(self, segments: List[dict]) -> List[dict]:
352
  """
353
+ Remove too long pauses in a segment by splitting the segment in two segments and removing the filled pause.
354
 
355
  Example (assuming (...) is longer than max_pause_length):
356
  [["Hello", "(...)", "World!"]] --> [["Hello"], ["World!"]]
 
376
  return split_segments
377
 
378
  @staticmethod
379
+ def _filter_segments_by_duration(segments: List[dict], min_duration: float = None, max_duration: float = None, ):
380
  """
381
  Removes segments with invalid duration
382
  :param min_duration: The minimum duration allowed for a segment.
 
394
 
395
  return filtered_segments
396
 
397
+ def _split_long_segments(self, segments: List[dict]) -> List[dict]:
398
  """
399
  Splits segments into smaller chunks if their duration exceeds the maximum chunk length specified in the config.
400
 
 
443
 
444
  return list_of_chunks
445
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
446